Developer tools now have a second audience: the coding agents using them on behalf of people. David Mokos explains what @expo learned when it started designing its platform for both human developers and AI agents.

Trying to force an LLM into perfectly deterministic behavior misses what it does best. Jakub Kuzimski shows how Allegro built a software factory around structured specifications, specialized agents, and explicit decisions instead.

A general-purpose agent can take on more than isolated coding tasks. Nick Miller shows how Grok Bot uses a cloud computer, company context, and familiar tools to handle long-running work across research, communication, and software delivery.

Agents can execute the same prompt-based skill again and again, spending tokens each time. Kent C. Dodds argues that repeated work belongs in deterministic software and demonstrates Kody, a personal software ecosystem designed to let any agent discover and run reusable capabilities.

Software development is moving from hand-written implementation toward systems where engineers direct agents, shape context, and judge the result. In the Agent Conf 2026 keynote, Mike Grabowski and Michał Pierzchała share how @callstackio is adapting to that shift. Follow @apex__ai and @agent_device for real-time updates.

Last week, our colleagues Andoni Santamaria and Felipe Ugena attended @AgentConf in Warsaw. 🔹 Specialized models 🔹 Multi-agent coordination 🔹 Observability 🔹 Production-ready agentic systems and more! @callstackio plainconcepts.com/agent-conf…

Levan’s talk is now ou YT

Had a great time at @AgentConf with @svennergr talking about how @grafana’s Agent Observability can help us, and everyone else building agents 🫵, build agents they can trust. The video is out now! 🎥👇 piped.video/XvgprsJDo3o?si=AqmC…

🍿

How does quantization affect APEX performance? 🔬 Artur Morys-Magiera @artus9033 and I shared our research at @AgentConf . We hope our findings help the community make more informed decisions about quantization. 🎥 piped.video/XLoY8PpYXNU

All talks from @AgentConf are live on YouTube. Here's the full playlist: piped.video/playlist?list=PL…

I talked at @AgentConf 2026 last Friday. Six years of harness engineering, trying to push models beyond their capability, gives us a bit of insight of where things will go next. In the end of the talk, I attempt to answer how we'll prompt AGI based on where we're heading right now. The talk is now available on YouTube, enjoy :) piped.video/watch?v=63_KrfUY…

Agent Conf is now on YouTube. Every speaker. Every talk. Ready to watch. A huge thank you to everyone who took our stage and gave the community new ideas to test and conversations to continue. Hit play 👇 clstk.com/4Axb1P0

You can run a Terra-capable model with ~17GB RAM. Measured on Apex on DGX Spark, @LeSiOO and @artus9033 got 3.4x decode speed at 27% of the initial model size with IQ4_XS quantization. At 99.6% mean normalized MMLU and tool use score, and 92% token agreement, I'd take it.

What would you build if you stopped worrying about tokens? In the @AgentConf keynote: how we went from chatting with agents to running software factories. We train, evaluate, and self-host our model Apex for a team of 200+. SOTA on React Native. Used daily for everything else.

My talk Build an Agent with Eve: The Open Agent Framework from AgentConf is available to watch: piped.video/watch?v=uVRmSOpi…

Agent Conf wrapped up on Friday. Now we're back with the Day 2 recap: agent fleets, software factories, state machines, observability, production monitoring, and "the end of apps." With @p_syche_, @lukasz_app , and @lukebfarrell 🔽 piped.video/h0gOMcEJbiU

344 runs. 25 days. Quantization is humbling 😅 Apex-31B BF16: 57% → 86.5% MMLU after fixing the chat template + eval harness. Same weights. Same 400 questions. 🤯 Then the payoff: IQ4_XS gave us 3.4× decode throughput and a 73% smaller file vs BF16 on DGX Spark. 🚀 @artus9033 and I talked about all of this at @AgentConf , organized by @callstackio. For me, it’s already one of the most exciting AI conferences in Poland! 🇵🇱🔥

Demo + small XState library coming soon (traveling back from @AgentConf where they got an early demo preview)